Implementation Assessment of Unpaved Road Condition

with High-Resolution Aerial Remote Sensing

Colin N. Brooks, Michigan Tech Research Institute (MTRI)

Dr. Tim Colling, P.E., Michigan Tech Center for Technology and Training (CTT)
Christopher Roussi, MTRI
Caesar Singh, P.E., US Department of Transportation Research & Innovative Technology
Administration (RITA)
David Dean (MTRI)
Richard Dobson (MTRI)
Dr. Melanie Kueber Watkins (CTT)
www.mtri.org/unpaved
RITARS-11-H-MTU1

www.mtri.org
 Characterization of Unpaved Road Conditions
through the Use of Remote Sensing

Goal of the Project: Extend available Commercial Remote Sensing & Spatial
Information (CRS&SI) tools to enhance & develop an unpaved road assessment
system by developing a sensor for, & demonstrating the utility of remote sensing
platform(s) for unpaved road assessment.
Commercially viable in that it can measure inventory and distress data at a rate and cost
competitive with traditional methods
Rapid ID & characterization of unpaved roads
Inventory level with meaningful metrics
Develop a sensor for, & demonstrate the utility of remote sensing platform(s) for
unpaved road assessment
Platform could be a typical manned fixed-wing aircraft, UAV, or both; depends on
relative strengths & weaknesses in meeting user community requirements
Simplify mission planning, control of sensor system, & data processing fitting for a
commercial entity or large transportation agency
Demonstrate prototype system(s) to stakeholders for potential implementation
developed through best engineering practices
Develop a decision support system to aid the user in asset management and planning

2
 Road Characteristics
Unpaved roads have common characteristics
Surface type
Surface width
Collected every 10', with a precision of +/- 4
Cross Section (Loss of Crown)
Facilitates drainage, typically 2% - 4% (up to 6%) vertical change, sloping
away from the centerline to the edge
Measure the profile every 10' along the road direction, able to detect a
1% change across a 9'-wide lane
Potholes
<1', 1'-2', 2'-3', >3 width bins
<2, 2-4, >4 depth bins
Ruts
Detect features >5, >10' in length, precision +/-2
Corrugations (washboarding)
Classify by depth to a precision of +/-1
<1, 1-3, >3
Report total area of the reporting segment affected
Roadside Drainage
System should be able to measure ditch bottom relative to road surface
within +/-2, if >6
Detect the presence of water, elevation +/-2, width +/-4 3

Float aggregate (berms)

 Combined Methods: Dept. Army
Unsurfaced Road Condition Index (URCI)

Representative Sample Segments (approx. 100 long; 2 per ~mile for

representative sample pg. 2-3 in TM 5-626)
2 Part Rating System
Density
Percentage of the sample area
Severity
Low
Medium
High
Good candidate method to focus on because
it offered a clear set of measurement
requirements, the realistic possibility of
collecting most of the condition indicator
parameters, and the potential applicability to
a wide variety of U.S. unpaved roads.
Endorsed by TAC as effective rating system
4
 Combined Methods: Dept. Army
Unsurfaced Road Condition Index

Decision matrix from distress criteria (Eaton 1987a)

5
 Summary of requirements
Number Name Type Definition
1 Data Collection Rate Sensor The systems must collect data at a rate that is competitive
with current practice (to be determined, TBD)
2 Data Output Rate System Processed outputs from the system will be available no
later than 5 days after collection
3 Sensor Operation Sensor easy, little training required
4 Platform Operation Platform Training needed TBD, based on platform choice
5 Reporting Segment System <100ft x 70ft, with location precision of 10ft. Map position
accuracy +/- 40ft
6 Sample locations System Specified by the user a map waypoints
7 Inventory System A classified inventory of road types is required prior to
system operation. This will consist of 3 classes: Paved,
Gravel, Unimproved Earth
8 Surface Width System This is part of the inventory, and may also be estimated by
the system measured every 10ft, precision of +/- 4
9 Cross Section Distress Estimate every 10ft, able to detect 1 elevation change in
9, from center to edge.
10 Potholes Distress Detect hole width >6, precision +/-4, hole depth >4,
precision +/-2. Report in 4 classes: <1, 1-2, 2-3, >3
11 Ruts Distress Detect >5 wide x 10 long, precision +/-2
12 Corrugations Distress Detect spacing perpendicular to direction of travel >8 -
<40, amplitude >1. Report 3 classes: <1, 1-3, >3.
Report total surface area of the reporting segment
exhibiting these features
13 Roadside Drainage Distress Detect depth >6 from pavement bottom, precision +/-2,
every 10ft. Sense presence of standing water, elevation
precision +/-2, width precision +/-4
14 Loose Aggregate Distress Detect berms in less-traveled part of lane, elevation
precision +/-2, width +/-4
15 Dust Distress Optional measure opacity and settling time of plume
generated by pilot vehicle
6
 Inventory: Surface Type
How many miles of unpaved road are there? Not all counties have this.
Need to able to determine this inventory
c. 43,000 (1984 estimate) but no up-to-date, accurate state inventory exists
c. 800 miles in Oakland County estimate
We are extracting this from recent, high-resolution aerial imagery, focusing on
unincorporated areas attribute existing state Framework roads layer
Completed Oakland, Monroe, Livingston, St. Clair, Macomb, Washtenaw,
Counties; shared with SEMCOG, adding to RoadSoft GIS asset management
tool
87%-94% accuracy
Ex: Livingston Co.: 894 miles unpaved
1289 miles unpaved

7
Unpaved Road Detection Results

Monroe County
Accuracy Assessment
at 30% coverage
Users Producers Overall

Unpaved 93.9% 77.5% 94.3%

Paved 94.3% 98.7%

Mileage
Paved 1390.0
Unpaved 351.9
Total
Mileage 1741.9
Integration of unpaved road inventory
results with RoadSoft GIS

9
Unpaved Road Detection Results

Oakland County
Accuracy Assessment
at 25% coverage

Users Producers Overall

Unpaved 83.6% 62.2% 89.4%

Paved 90.5% 96.7%

Mileage
Paved 2948.2
Unpaved 693.9
Total
Mileage 3642.1
Unpaved Road Detection Results

Macomb County
Accuracy Assessment
20% coverage
Users Producers Overall
Unpaved 71.8% 60.9% 94.3%
Paved 96.2% 97.6%

Mileage
Paved 1847.0
Unpaved 319.4
Total
Mileage 2166.4
Unpaved Road Detection Results

Livingston County
Accuracy Assessment
25% coverage
Users Producers Overall

Unpaved 83.8% 72.1% 87.2%

Paved 88.4% 93.8%

Mileage
Paved 1289.4
Unpaved 894.1
Total
Mileage 2183.5
 Selected sensor: Nikon D800
Body type
Body type Mid-size SLR

Nikon D800 full-sized (FX) sensor, 36.3 Mp, Body material

Sensor
Magnesium alloy

4 fps - $3,000 Max resolution

(px)
7360 x 4912

Effective pixels 36.3 megapixels

More than meets all our requirements Sensor photo

detectors
36.8 megapixels

Other resolutions 6144 x 4912, 6144 x 4080, 5520 x 3680, 4800 x 3200, 4608 x 3680, 4608 x

Weight prime lens, weights ~1.5 kg Image ratio w:h

3056, 3680 x 2456, 3600 x 2400, 3072 x 2456, 3072 x 2040, 2400 x 1600
5:4, 3:2
Sensor size Full frame (35.9 x 24 mm)
Sensor type CMOS
Processor Expeed 3
Color space sRGB, Adobe RGB
Color filter array Primary Color Filter
Image
ISO 100 - 6400 in 1, 1/2 or 1/3 EV steps (50 - 25600 with boost)
White balance 12
presets
Custom white Yes (5)
balance
Image No
stabilization
Uncompressed .NEF (RAW)
format
JPEG quality levels Fine, Normal, Basic
File format NEF (RAW): 12 or 14 bit, lossless compressed, compressed or
uncompressed
TIFF (RGB)
JPEG
Optics & Focus
Autofocus Phase Detect
Multi-area
Selective single-point
Tracking
Single
Continuous
Face Detection
Live View
13
 Platforms
Bergen Helicopter
Total flight time: 16 minutes (not including 2
minute reserve); flight time for a 200 meter
section ~ 4 minutes.
Flown at 2 m/s at 25 and 30 meters
50mm prime lens

Cessna 172 and 152 Aircraft

Average air speed: 65 knots (~ 75 mph)
Flown at altitudes of 500 and 1000 feet
105 mm prime lens (2012), 70-200mm zoom
(2013)

Bergen Hexacopter
Total flight time: up to 30 minutes with small
payloads
Weight: 4kg unloaded
Maximum Payload: 5kg
Includes autopilot system, stabilized mount
that is independent of platform movement,
and first person viewer system (altitude,
speed, battery life, etc.)
14
 Initial UAV Collect

Flight time for a 200 m section: 4 minutes

During collects helicopter is flown at 2 m/s
and at an altitude of 25 m (82) and 30 m
(98)

Example flight at http://www.youtube.com/watch?v=KBNQzM7xGQo 15

Field site collections
Five sites were selected in 2012, four
sites were selected in 2013 in SE
Michigan
Assistance of Road Commission Authorities
aided in the selection of field sites
None of the sites contained all distress
features of interest for ground truth
assessment, but all were found
Road graders often hindered data collection

Two collections opportunities in Iowa

and Nebraska (August 2013)
Verified maintained roads (with the potential
of being maintained using different materials
and methods) in other states could be
categorized with the same processing suite as
Michigan roads
Selections based on Google Earth imagery
and proximity to Interstate-80
Results indicate that there were no issues in
assessing road conditions on these other
unpaved roads.
16
Ground Truth

17
Helicopter Data Garno Rd.
25m Altitude
 Performance Collected Imagery

Taken from 25m

19
Aerial Data Piotter Rd.
500 ft Altitude
 3D Reconstruction (Helicopter)

Initial point cloud Densified point cloud

3D surface from point cloud

21
3D of Piotter Rd (Hexacopter, 27 images)

22
3D of an Iowa Road (Hexacopter, 18
images)

23
 3D data examples
Important to categorizing distresses by severity
Obtaining 0.9 cm ground sample distance

24
 Distress Detection Potholes

Canny Edge detection used to locate edges

Hough Circle Transform is used to locate potholes
Edge Detection Identified circles

Note: Circles near edges ignored. 25

Distress Detection Washboarding

26
 Distress Detection Washboarding

Missing
due to
area
threshold

Ground Truth Corrugation Area: Computed Corrugation Area:

19.6 sq. m 17.2 sq. m
27
 Algorithm Performance Summary

In summary, the following data collection parameters will meet all

system performance requirements:
24M-36M-pixel sensor
50mm, f/1.4 lens set at f/2.8
1/250s (maximum) shutter speed (shorter is better)
ISO set as needed for proper exposure given ambient lighting
Distance of 20m-30m from surface
2m/s (maximum) forward speed
2fps (minimum) image capture rate (obtained with a simple
intervalometer)
64GB high-speed storage medium
Results from this system - User feedback: results appearing useful,
implementation needed
The Asset Management Council of Michigan, Southeastern Michigan Council of
Governments, Road Commission for Oakland County; sharing results with South
Dakota DOT 28
 Algorithm Performance Summary

Pothole:

Crown Damage:

Rut Detection:

Corrugation Detection:

29
 Aerial Sensor Performance
Algorithm performance, and the ability to meet the stringent requirements on resolution,
depends on the ability to collect data that has enough angular diversity to be able to
reconstruct three dimensions from two dimensions.
As the distance from the ground increases, the solid angle that any object subtends decreases, and
at some point, becomes too small for high-resolution reconstruction.
Data taken from an altitude of 500 feet do not meet the system requirements in resolution. That is,
the reconstructed pixels have been found to be too large. This is due to the lack of sufficient
angular diversity.
Solutions:
More data are collected with the camera points at the
same point on the ground, but at oblique (as well as
nadir) views.
Several passes over the same location can be made,
with the camera at different angles.
Much higher resolution sensors, with a wider-angle lens
than the 200mm currently used, would allow data to be
taken in a single pass.

Use of a sensor at altitudes above 400 feet is not

practical at this time, only sensors flown at altitudes
below 100m will meet all the performance (i.e.
resolution) and cost-effectiveness requirements.

30
 Analyzed data are integrated into
RoadSoft GIS Decision Support System

31
 Road Analysis Process Flow RoadSoft DSS
integration

3 Surface Identification 2
1
Identify Data From Images
Collect Aerial
Unsurfaced Aerial Imagery
Imagery
Road Network Analysis

Surface Identification
Manual Inspection Functions in
7
Compile
eCognition
Distress Data
Distress and From Platform
Inventory Data 5
For Samples 4
Fly Data
6 Identify Sample
Collection
Data Processing Locations In
Distress Data From Sorties with
Flight System
Manual Inspection Platform

8
Assign Samples Functions in
to Represent Network Field Report
Network Condition Platform System
Report

9 10 12
11
DSS Analysis of Selection of Determine Data
Record
Data Candidates & Needs and Repeat
Competed Work
Scheduling Cycle

Maintenance Completed Functions in

Plan & Budget Project History
RoadSoft 32
DSS Ranking System

33
 Costs Manual Characterization

Cost assumptions are described in detail in Deliverable

7-B that will be posted to the project website once
approved.

34
 Costs Remote Sensing
UAS (UAV, high-resolution camera, and good-quality lens):
Cost per mile rated $30,590/yr/1575 mi/yr = $19.42/mi rated.
HOWEVERtwo 100-foot measured segments represent one mile of
road, so 5,280 ft/200ft is 26.4. Therefore each mile of measured road
represents a road network 26 times larger.
Therefore cost is $0.74 per mile, in addition to the cost of vehicle use
($0.55/mi)
8 hours/day, 3 days/week, 21 week season to collect 300 road-miles of data segments

Manned Fixed Wing:

Cost per mile rated $54.47 per mile assessed for up to five sites
per mile
$10.26 per mile (generous assumption of continuous data
collection)
$16,340 for same type of analysis as listed above
Caution must be made for cost comparisons between remote sensing and
manual characterization of road conditions due to the resolutions of the
outputs; centimeter-by-centimeter analysis of entire road segments is
essentially impossible via manual inspection.
35
 Administrative Issues FAA regulations
It should be noted that current (as of October 2013)
FAA regulations do not adequately address UAS
operations for private entities.
The FAA document 14 CFR Part 91
(http://www.faa.gov/about/initiatives/uas/reg/media/frnotice_uas.pdf
) specifically excludes individuals or companies flying model aircraft
for business (commercial) purposes.
For public entities (such as the USDOT), the process of operating a
UAS involves obtaining a Certificate of Authorization (COA) for a
particular mission. Each mission must have its own COA, which
effectively prevents the current use of UASs for arbitrary unpaved
road assessment. Thus, under current FAA guidelines, there is no
way to deploy an unmanned system for this purpose.
However, some agencies with COAs have been able to get them
reapproved within relatively short time periods (< 1 month).
New Dec. 2013 5-year FAA UAV integration RoadMap

This may change by late 2015, when the FAA has to

have established regulations dealing with Unmanned
Aerial Systems (UASs) in the National Airspace
System (NAS).
New regulations for small UAVs (SUAS) due by Nov.
2014 file & fly for under 55 lbs SUAS?
More practical deployment starting in 2015 -
commercially 36
 Contact Info
Colin Brooks cnbrooks@mtu.edu
Desk: 734-913-6858, Mobile: 734-604-4196
Michigan Tech Research Institute, MTRI
3600 Green Court, Suite 100
Ann Arbor, MI 48105
www.mtri.org
Tim Colling, Ph.D., P.E. tkcollin@mtu.edu
Chris Roussi croussi@mtu.edu
Rick Dobson rjdobson@mtu.edu
David Dean dbdean@mtu.edu
Melanie Keuber Watkins, Ph.D., P.E. mkueber@mtu.edu
www.mtri.org/unpaved

DISCLAIMER: The views, opinions, findings and conclusions reflected in this

presentation are the responsibility of the authors only and do not represent the
official policy or position of the USDOT/RITA, or any State or other entity.
37